About the talk
An end to end deployment of an ML model using Azure ML, Azure Kubernetes Service, Application Insights
NASA Citizen Scientist; TCS Utilities Data and Analytics Practice Lead and Energy Trading and Risk Management Practice LeadView the profile
Play? What I'll do is chat. I'd speak about as you. Do you let me share. What's the story so far? I guess just say that, I have competitions science competition. So I'm referring to this data science competition for that model and the competition ranpur around Pumas shoes. And there is so many different models. I'm doing different hyperparameters, Model Management. In this case of things with your being better and scored a soda experiment. Similarly while writing a research paper for this cancer images classification. So
I can finish the report as bad of me and model parameter how to create models and that's why I need black pants like legit. And it's the magic Model Management park with some other place where we can do all these things in the And then we will actually be deployed and then we'll go from. There is a model train days, amadi incidents, and bodies. And then there is or are we trying to share the end of the session? So if you do not follow anything, don't worry is explain. Why am I dreaming
results explained listening? Enjoy and ask me any questions? Dial, move away from this place. And the good thing is, I just gave a station for a 0°, and I prepared the flight, but I gave it to the notebook to swing by the notebook and a location subscription ID and the Tenant. So this is what it. What it means is that the workspace is something like It is a home for everything. So, this is what happened. What happened to stay to work spaces, created application, in science, is created and
storage account, is created space, all these things as you created for me. So you can see it wants you to excuse the poor. All these things are being created. Then what is to be done? Now, considered the word space, like a house or intention? Sorry. I didn't say that. All intention is to predict the quality of wine given different job to do. So. Do you have to be so Divine letter? It's actually pretty interesting problem. I don't know whether or not important thing because it gives you about what the
price of the wine would be our problem instead of giving these things that you do. And then we last. We are trying to build a modern and what we do, we upload the data before the load the data. What it will do. I go back to my screen out here. So this is storage account which is associated and in the storage account, that is a container. And what it does is set for datastore and it will create a fine over here. I don't know what we're done till now, repeated the words space.
So here is a fancy thing. What are you is chatting. The training folder basically just need a folder. Weingold what it means. It's dead here in my world. Ain't I guess. One folder call winebow in what I do is chat now. So you have do you have a lot of space? What do you need a machine right around where the model will be so far. Dad gives me a case in which I can be a cluster of machines. So it's hard to compute cluster and I can for this is for me, a very simple thing. So I will just be the
one note. In the computer section, so I ran it before this meeting. So computer class. Sure. He's already created you. So this is standard be too. If you had with me, I needed the work space. I've uploaded the data needed the class. If not, what I do is that. So basically this is my mom. So I'm using YouTube am and the data folder and the standard position 380 and all this stuff in space because this is the only piece you can't do anything. I need to do is honor his decision to evict you. They said you can do anything
because you're still alive, but now what happens, if an environment you would require that you have built a model and You have some, you know, required of an environment in which you can. Eraser able to see my see, my screen, right? Hello. Hello, you are able to see my screen, right? I guess, Jacob course, okie dokie pokie where they go back to. So it's so worth the wait was like so he created the cluster and now we created the environment and what we do with that
Louise, Steadman so they can see, m is what it does. Is that it thinks the training split. It takes her parents. It takes a computer with the registered environment and combines all of this together. So this is the estimated for me. So what so what we'll do is create an experiment experiment together and we You can see the experiment in the experiment stand. So, this experiment with the information between have the signature? So so what would happen instead? We will have the signature and the estimated
experiment out you. So the experiment has all the ranks so we can stop me to go downstairs. When you submitted on what will happen is that he now the model we get executed. So here, once you can see these experiments. There are different types of run, some errands. So I know I've got one experiment and I can change my pm and I can do a number of advanced. So these other errands. So you see here, this message if I didn't intend and then I did a and then I did
the RMS TV to see where all this is that you can stop and all this is being lured into this experiment. In the workspace one experiment once everything is in 1. Pint. Do you think the data store created, the modeling speed, created the computer cluster, created the computer cluster? You see the computer cluster, read it an experiment. And so in the cold I have just put the only one run you can put as many Iranians be time so you can get it, free Lance around sport for. So that's how all these things. That no one thing had the model. I can register the model in the workspace.
What it means is that now my model is in a very secure base and now I can't use it in a number of times. Good morning is associated with the r on Adidas Adidas, Gore-Tex, everything. So this is my model and Bill this model and put it in that space. All these marbles. I do not have to peel tomatoes. So what are you do, you do weather data scientist? Inference, inference more inferencing part of it? Instant imprints screen is basically a place in which you take the data
and you are going to pray. So what are you going to take the data? Say, for example, you are giving the data has things, right? You're giving the data and please fix the acidity will tile acidity citric, acid and you're going straight to why you are getting the model Divine model, which is there in the workspace, you loading them out and once you take the data, which is coming in and then you pretty Now this model is done. Now. What you do is that you have to pay to Independence, because
this is, why are you psychic site appear? Dependency. What it means is that what does Creek you have their dependency any combined together? You said, once you have all these things now, you need to create an inference. Closter Closter, because this is so, so here you may not be quite a plastic, but it's supposed to be in production. It would require mini requested, So you would require You can't deploy you so bored. And plus I have to wait when I run this code. I get a plaster. Which if you go here, you compute. You have the computer yesterday
for the model building, and then you have an inference clusters. Listen to Prince cluster. So late is coming. But you have, do you have the screen? You have been couldn't sleep. And then what you do is chat, you divine mother. Okay, you deploy the Irvine model. The wine model is applied with a model here. So it's not a problem for me. So all this is said and so what is the inference conflict? It also requires a switch. We had already done the exact same inference. Auntie Dee is here and
production, which is So once it is done, so what we do is chat. We want to get into point of view, taken the requests from people in areas. Are you taking from PSU email? So what are you do? Is chat me. And important point which is created out here. Don't be confirm. If you see here is the endpoint stage should be held. Otherwise, you can so you can see the different types of. Dell model, which has been deployed in the plaster is also bought an application inside. So now what is Aspen? You can't,
we have a piece service and which we have deployed with an endpoint of the same point is actually, the same point, is said, he cannot utilize it to call and see improving the quality of life and wants to get the keys, then I'm passing this data. So there's data, is this okay? Arrangement is data and what I'm going to predict this this quality of the. So what I'll do with that. So this is all the detail. It is coming in, not hear the primary key, which is, which is a key beside, his thing is coming and the response response
because the quality of the wine is telling what we're done till. Now, I've got everything which you can see here. You can Google application insights from here. What you can do is set you to hear this application in science. And application insights that has the logs. Okay, and you can query application insights or but the predictions. Inside the queen and a production system is needed to work space. We created an inference plaster and then eat and then recreated. So all this
is is in one place. All of it. Which, which we have used to work space. You see the computer cluster, which was used for for modeling input. You see the endpoints also between the models, which we have been registered by planes, are not used any pipeline pipeline. Experiment experiment is, it's a place where you can have a number for me. He lands and bring as many runs and white Jameson, see all these parameters and I can roll all these drawings. So I can build different different models and everything is. Maybe you can just download it
or you can create a space and you can have all of it. And what of our production I do not have to worry. How will I deployed in production? Everything is taken care of? So he doesn't come with as your email, it comes with integration, with keyboard with the with the with the insides are being very honest or a scientist. This is something. So this is what I have.
Buy this talk
Buy this video
Our other topics
With ConferenceCast.tv, you get access to our library of the world's best conference talks.